Thiamine (B1) cycling related protein HMMs and metagenome assembled genomes

Sequence and data analysis Data from seven metagenomic samples sequenced herein were processed as follows. Illumina raw reads were quality filtered using Trimmomatic (v.0.39), trimming after a base with a quality score below 3 or when the 25 bp moving average was <30. Reads were excluded from further analysis when <50 bp after trimming. On average, 44,341,428 ± 3,337,198 quality-controlled paired-end reads were obtained from the Illumina NovaSeq platform. Samples were assembled separately using megahit (v.1.2.9; [44]) with the ‘—presets meta-sensitive’ option and considering both paired and unpaired reads that passed quality control. The assembly was filtered using the ANVI’O script ‘anvi-script-reformat-fasta’ (v.7), and only contigs longer than 5 Kb were retained for future analysis. Prodigal (v2.6.3) was used for protein prediction, and functional annotations were obtained using prokka (v1.12). In addition, the blastKOALA and ghostKOALA tool servers were used to obtain KO annotations for predicted proteins. Functional annotations were generated using hmmscan (v.3.3.2; Table S1) with e-value <1^-10 vs PFAM (v.35) and eggNOG-mapper (v.2.0.1) with the ‘-diamond’ flag. For predicted proteins, hmmsearch was run with additional HMM profiles (B1-related biosynthesis, salvage, and transporter proteins) obtained from TIGRFAM (v15.0) and Paerl et al. 2018 (Table S1, Supplementary Data S12).Metagenomic bins were constructed using all versus all samples on MetaBAT2 (v2.12.1 (46)), based on tetranucleotide frequencies and differential coverage from sequence alignment maps generated with Bowtie2 (v2.4.1 (47)) using default settings and considering only paired-end reads for mapping to assemblies. Sequences that could not be assigned to a bin were retained for downstream analysis and annotation.